Experiments with Self-Tuning Control of Flotation

Abstract

This paper describes initial experimental development of a self-tuning regulator to control the cleaner section of a zinc flotation circuit in a Canadian concentrator. As the zinc content of feed ore to the circuit varies, circulating loads in the cleaner section vary markedly, altering process dynamics sufficiently to degrade the performance of the proportional-integral regulator currently in use, even resulting in instability. The extremely complex dynamics of the circuit, and the lack of reliable dynamic models for flotation, preclude the use of simulation for design. It was thus necessary to develop the controller entirely by plant experiment. Random fluctuations in the plant were extremely high. A combination of step tests and least-squares parameter estimation runs was used to establish a low-order model in differenced variables. Parameter estimates were found to converge, although the variance of the estimates were relatively high owing to the high process noise. A minimum-variance controller using the continuously estimated parameters of the model was then implemented. In practical implementation of the controller, it was found necessary to add supplementary integral action to reduce offsets, and to bound the control variable and place a lower limit on estimated process gain to ensure stability during startup. Storage and computing time requirements for the self-tuning controller were acceptable. Preliminary experimental results indicate that the self-tuning regulator remains stable under all conditions tested, unlike the conventional controller However, auto-correlation analysis of the residual error indicates that the model used fails to include significant disturbance effects. Nevertheless, it is concluded that despite the high costs of plant experiments potential improvements in process operation warrant continued development.